Hematology plays a critical role in diagnosing and managing a wide range of blood-related disorders. The manual interpretation of blood smear images, however, is time-consuming and highly dependent on expert availability. Moreover, it is particularly challenging in remote and resource-limited settings. In this study, we present an AI-driven system for automated blood cell anomaly detection, combining computer vision and machine learning models to support efficient diagnostics in hematology and telehealth contexts. Our architecture integrates segmentation (YOLOv11), classification (ResNet50), transfer learning, and zero-shot learning to identify and categorize cell types and abnormalities from blood smear images. Evaluated on real annotated samples, the system achieved high performance, with a precision of 0.98, recall of 0.99, and F1 score of 0.98. These results highlight the potential of the proposed system to enhance remote diagnostic capabilities and support clinical decision making in underserved regions.

This study introduces a novel framework leveraging Rough Set Theory (RST)-based feature selection—MLReduct, MLSpecialReduct, and MLFuzzyRoughSet—to enhance machine learning performance on uncertain data. Applied to a private cardiovascular dataset, our MLSpecialReduct algorithm achieves a peak Random Forest accuracy of 0.99 (versus 0.85 without feature selection), while MLFuzzyRoughSet improves accuracy to 0.83, surpassing our MLVarianceThreshold (0.72–0.77), an adaptation of the traditional VarianceThreshold method. We integrate these RST techniques with preprocessing (discretization, normalization, encoding) and compare them against traditional approaches across classifiers like Random Forest and Naive Bayes. The results underscore RST’s edge in accuracy, efficiency, and interpretability, with MLSpecialReduct leading in minimal attribute reduction. Against baseline classifiers without feature selection and MLVarianceThreshold, our framework delivers significant improvements, establishing RST as a vital tool for explainable AI (XAI) in healthcare diagnostics and IoT systems. These findings open avenues for future hybrid RST-ML models, providing a robust, interpretable solution for complex data challenges.

Pain estimation is a critical aspect of healthcare, particularly for patients who are unable to communicate discomfort effectively. The traditional methods, such as self-reporting or observational scales, are subjective and prone to bias. This study proposes a novel system for non-invasive pain estimation using eye-tracking technology and advanced machine learning models. The methodology begins with preprocessing steps, including resizing, normalization, and data augmentation, to prepare high-quality input face images. DeepLabV3+ is employed for the precise segmentation of the eye and face regions, achieving 95% accuracy. Feature extraction is performed using VGG16, capturing key metrics such as pupil size, blink rate, and saccade velocity. Multiple machine learning models, including Random Forest, SVM, MLP, XGBoost, and NGBoost, are trained on the extracted features. XGBoost achieves the highest classification accuracy of 99.5%, demonstrating its robustness for pain level classification on a scale from 0 to 5. The feature analysis using SHAP values reveals that pupil size and blink rate contribute most to the predictions, with SHAP contribution scores of 0.42 and 0.35, respectively. The loss curves for DeepLabV3+ confirm rapid convergence during training, ensuring reliable segmentation. This work highlights the transformative potential of combining eye-tracking data with machine learning for non-invasive pain estimation, with significant applications in healthcare, human–computer interaction, and assistive technologies.

Effective pain assessment in infants aged 0–3 months is a critical challenge in neonatal intensive care units (NICUs) and family medicine clinics, where self-reporting is impossible and current observational tools remain subjective and inconsistent. This paper presents a pilot study proposal for an AI-driven pain monitoring dashboard designed to support clinicians and caregivers in real time. The system integrates multimodal inputs, including facial video, cry audio, and optional physiological signals, processed through a deep learning framework with self-supervised pretraining and few-shot adaptation to overcome limited labeled neonatal data. A clinician-friendly dashboard displays continuous pain scores, highlights salient modality contributions, and issues alerts during high-risk events. The approach ensures transparency and trust by providing explainability through saliency maps and modality-specific visualizations. This pilot will evaluate the feasibility, usability, and clinical impact of the dashboard in NICU and family clinic environments. The proposed system offers a practical and ethically compliant pathway for deploying explainable AI in early-life pain management, with potential to improve care quality and support medical decision-making.

Effective pain assessment in infants aged 0–3 months is a critical challenge in neonatal intensive care units (NICUs) and family medicine clinics, where self-reporting is impossible and current observational tools remain subjective and inconsistent. This paper presents PANDIA (Personalized Adaptive Neuro-symbolic Data-fusion for Infant Assessment), a novel multimodal AI system that combines hierarchical representation learning, graph-based inter-modal reasoning, meta-learning personalization, and symbolic concept-bottleneck explanations for robust infant pain assessment. Unlike transformer-centric approaches, PANDIA employs lightweight CNN/TCN backbones with a graph neural network for inter-modal fusion, achieving clinical interpretability through explicit concept bottlenecks and symbolic reasoning. Our federated learning framework enables privacy-preserving multi-site collaboration while meta-learning adaptation provides personalized assessment with minimal per-infant data. Evaluated on 2,847 infants across four datasets, PANDIA achieves 87.3% accuracy with 92.1% clinician acceptance rate for explanations, achieving a 12.4% accuracy improvement over the best baseline, consistent across all four datasets and an independent out-of-distribution test set, while maintaining fewer than 30M parameters for edge deployment. The proposed system offers a structured and interpretable step toward deploying explainable AI in early-life pain management, with potential to improve care quality and support medical decision-making. Key limitations include the retrospective validation design, dataset heterogeneity across collection sites, and the need for prospective clinical trials before deployment in live clinical settings. All code, trained models, preprocessing pipelines, and supplementary materials are fully publicly available without restriction at: https://github.com/oussama123-ai/pandia . The NICU-MM dataset is available upon request subject to an ethical data use agreement; the access procedure is detailed in Section 4.1.1.

Pain assessment remains challenging for patients unable to verbally communicate, including neonates, cognitively impaired individuals, sedated patients, and those who suppress expressions due to cultural norms or stoicism. We demonstrate that integrating thermal imaging with RGB facial expression analysis provides more accurate and robust pain intensity estimation than either modality alone. Our dual-camera system records synchronized thermal and RGB video, processed through a cross-spectral attention fusion (CSAF) model with a temporal transformer for continuous 0–10 scale pain prediction. In a controlled laboratory pain induction study, 50 healthy adults (ages 21–68, 87.3 h video) underwent Cold Pressor Test and pressure algometry protocols; our system achieves MAE = 0.79, representing a 33.1% improvement over RGB-only (MAE = 1.18, p < 0.001). In a real-world clinical postoperative monitoring study, 30 surgical patients (ages 31–74, 17.7 h video) recovering from abdominal surgery were monitored; our system achieves MAE = 1.08, representing a 28.5% improvement over RGB-only (MAE = 1.51, p < 0.001). Across the combined cohort ( n = 80), MAE = 0.87 (29.3% overall improvement over RGB-only). Benefits increase at higher pain intensities (38.5% at severe pain) and for challenging populations where expressions are suppressed (37.6% for low expressers). Thermal responses precede visible expressions by 1.2 ± 0.3 seconds, enabling earlier detection. This work was validated on adults only; pediatric applications require dedicated validation. Translation to clinical practice requires multi-site prospective trials, regulatory approval, and careful implementation planning.